ThoroughWiz

The Record

The new record, kept the way the credible operators keep theirs: every book on one page, the count printed on every number, the market's own benchmark beside ours, and the ledger files underneath so anyone can recompute a figure. The record begins November 17, 2026 for racing and November 18, 2026 for sports — the fresh-start summary explains what the old record found and why it closed. Updated nightly after settlement. This page was built November 18, 2026, 1:42 PM ET.

Tier 1 — the recorded record

Every row is computed from the public ledger files linked below, from picks filed before the event. Windows: the last 30 days and the series to date. Series starts: the racing books begin November 17, 2026; the sports books begin November 18, 2026 — the reset date in the methods block below.

BookWindowSettledW–LWin %P&LROIBenchmark
Racing — the pick (Selection Protocol v3.19)Last 30 dayssmall sample2611–1542.3%$+1.06+2.0%ML favorite ROI -26.0% (same races, n=26)
Series to datesmall sample2611–1542.3%$+1.06+2.0%ML favorite ROI -26.0% (same races, n=26)
Racing — Top Four seriesLast 30 dayssmall sample2620–676.9%$-64.92-32.5%ML favorite ROI -26.0% (same races, n=26)
Series to datesmall sample2620–676.9%$-64.92-32.5%ML favorite ROI -26.0% (same races, n=26)
NFL (Rule v1.9)No new-series games settled yet — Week 11 opens November 19.Last 30 days0—————
Series to date0—————
NCAAB (Other-Sports Rule v2.1)No settled games yet in the new series.Last 30 days0—————
Series to date0—————
Pick of the Day — RacingLast 30 dayssmall sample11–0100.0%$+0.40+20.0%—
Series to datesmall sample11–0100.0%$+0.40+20.0%—
Pick of the Day — NFLNo settled cards yet in the new series.Last 30 days0—————
Series to date0—————
Pick of the Day — NCAABNo settled cards yet in the new series.Last 30 days0—————
Series to date0—————
Syndicate — Pace (racing)Last 30 dayssmall sample85–362.5%$+16.60+103.8%ML favorite ROI -23.4% (same races, n=8)
Series to datesmall sample85–362.5%$+16.60+103.8%ML favorite ROI -23.4% (same races, n=8)
Syndicate — Connections (racing)Last 30 dayssmall sample80–80.0%$-16.00-100.0%ML favorite ROI -70.9% (same races, n=8)
Series to datesmall sample80–80.0%$-16.00-100.0%ML favorite ROI -70.9% (same races, n=8)
Syndicate — Price (racing)Last 30 dayssmall sample72–528.6%$-6.54-46.7%ML favorite ROI +7.6% (same races, n=7)
Series to datesmall sample72–528.6%$-6.54-46.7%ML favorite ROI +7.6% (same races, n=7)
Syndicate — Contrarian (racing)Last 30 dayssmall sample82–625.0%$+5.36+33.5%ML favorite ROI -85.0% (same races, n=8)
Series to datesmall sample82–625.0%$+5.36+33.5%ML favorite ROI -85.0% (same races, n=8)
Syndicate — Lone Wolf Book (racing)Last 30 dayssmall sample72–528.6%$-7.00-50.0%ML favorite ROI -80.0% (same races, n=7)
Series to datesmall sample72–528.6%$-7.00-50.0%ML favorite ROI -80.0% (same races, n=7)
Syndicate — Pace (sports)The November 17 edition settles on the morning of November 18.Last 30 days0—————
Series to date0—————
Syndicate — Connections (sports)The November 17 edition settles on the morning of November 18.Last 30 days0—————
Series to date0—————
Syndicate — Price (sports)The November 17 edition settles on the morning of November 18.Last 30 days0—————
Series to date0—————
Syndicate — Contrarian (sports)The November 17 edition settles on the morning of November 18.Last 30 days0—————
Series to date0—————
Syndicate — Lone Wolf Book (sports)The November 17 edition settles on the morning of November 18.Last 30 days0—————
Series to date0—————
All books totalFlat-stake units ($2 = 1 unit for the dollar books); a play filed in two books — Pick of the Day and its league book, or a reader book and the Lone Wolf Book — counts once, and the same sports bet backed by several Syndicate readers counts onceSeries to date8340–4348.2%-32.22u-20.5%—

P&L basis: racing books $2 flat to win; sports books 1 unit flat at the filed price; Syndicate books at their filed $2 flat stakes. Rows under 30 settled carry a small-sample flag. Benchmarks: for racing, the morning-line favorite's win ROI on the same settled races; for sports, closing line value where a closing capture exists.

Racing pick book — race by race

Every settled race in the Tier 1 pick book, from the same ledger rows the totals above are computed from. Stakes are $2 flat to win. Filter by track or surface — the count and record below recompute from the rows shown.

DateRaceTrackDistanceSurfacePickMLResultReturned
2026-11-17ZIA_R9Zia Park5½ FurlongsDirt#8 Ragazza Vincente6/1loss$0.00
2026-11-17ZIA_R5Zia Park5 FurlongsDirt#5 Classic Moment7/5win$2.80
2026-11-17ZIA_R4Zia Park6 FurlongsDirt#1 Amazing Music7/5win$4.80
2026-11-17ZIA_R1Zia Park5½ FurlongsDirt#4 Mother of Pearl3/1win$4.20
2026-11-17PRX_R9Parx Racing1 MileDirt#4 Rhee Wall7/2loss$0.00
2026-11-17PRX_R8Parx Racing6½ FurlongsDirt#3 Rock Me Gently5/2loss$0.00
2026-11-17PRX_R7Parx Racing1 MileDirt#1 Drum Roll Please9/5loss$0.00
2026-11-17PRX_R6Parx Racing1 MileDirt#6 Snookie Bear5/2loss$0.00
2026-11-17PRX_R5Parx Racing1m70ydsDirt#1 Jackson Road2/1loss$0.00
2026-11-17PRX_R4Parx Racing1 MileDirt#6 America's Honor2/1win$11.20
2026-11-17PRX_R2Parx Racing1 MileDirt#1 Peridot Pendant8/5win$2.80
2026-11-17MVR_R8Mahoning Valley6 FurlongsDirt#1 Cowgirl Chic5/2loss$0.00
2026-11-17MVR_R7Mahoning Valley6 FurlongsDirt#2 Reckless L J3/1loss$0.00
2026-11-17MVR_R5Mahoning Valley6 FurlongsDirt#2 Right On the Munny5/2loss$0.00
2026-11-17MVR_R3Mahoning Valley1 MileDirt#1 Moreau2/1win$4.20
2026-11-17MNR_R8Mountaineer6 FurlongsDirt#10 Fregon2/1loss$0.00
2026-11-17MNR_R7Mountaineer5½ FurlongsDirt#2 Back On Track2/1win$2.20
2026-11-17MNR_R5Mountaineer1 MileDirt#5 North Warden2/1win$7.60
2026-11-17MNR_R4Mountaineer5 FurlongsDirt#7 Yalina1/1win$2.40
2026-11-17MNR_R2Mountaineer5 FurlongsDirt#9 Just for the Money3/1win$6.20
2026-11-17MNR_R1Mountaineer1 MileDirt#4 Semper Anticus3/1loss$0.00
2026-11-17FL_R9Finger Lakes5½ FurlongsDirt#12 Escape Hall7/2loss$0.00
2026-11-17FL_R8Finger Lakes1m70ydsDirt#10 Probability3/1loss$0.00
2026-11-17FL_R6Finger Lakes5½ FurlongsDirt#2 Froghollowsummer5/2win$4.66
2026-11-17FL_R5Finger Lakes1m70ydsDirt#5 Tiz Goldy5/2loss$0.00
2026-11-17FL_R2Finger Lakes1m70ydsDirt#1 Good Cop8/5loss$0.00

The ledger files

The numbers on this page are computed from these files — one row per settled leg, with the date, the event, the pick, the filed price, the result and the return. Anyone can recompute every figure on the page from them.

How this record is kept

Staking basis. Racing books: $2 flat to win per leg (the Top Four stakes $2 on each of the four that start; a scratched member refunds its $2). Sports books: 1 unit flat at the filed price. The Syndicate books: their filed $2 flat stakes. Pick of the Day sports cards: 1 unit at the filed price; racing cards $2 flat. The all-books total converts each book to flat-stake units ($2 = 1 unit for the dollar books) before summing, and prints below the per-book rows. A play filed in two books — a Pick of the Day card and its league book, or a Syndicate reader book and the Lone Wolf Book — counts once in the total, and so does the same sports bet backed by several Syndicate readers: each reader's own book row stands as that reader's record, but the bet itself is one play.

The lock. Picks are filed and timestamped before the event, and a filed set never changes: a later build for the same date re-renders the filed set or refuses. Day pages freeze at settlement; corrections, if any are ever needed, print as dated notes, never silent edits.

Settlement sources. Racing settles against official Equibase charts. Sports settle against ESPN finals through the labs' own settlement passes. The Syndicate settles through its morning settlement pass. Pushes count in the settled total with the stake returned; voids are excluded from the settled total and counted where a book has them. Rows under 30 settled carry a small-sample flag.

Benchmarks. Racing rows carry the morning-line favorite's win ROI on the same settled races. Sports rows carry closing line value where a closing capture exists — the totals capture for the lab books, the Syndicate's own closing capture for its sports rows. Closing capture is partial: a book with no captured closes prints a dash, and no CLV is estimated.

Versions in force. Racing: Selection Protocol v3.19 from November 18, 2026 (one booked horse per race — the pick). Day one of this record — the November 17, 2026 card — was filed under Selection Protocol v3.18; its official picks are booked into this record from day one on the Lab Owner's order. From November 18 the book files under v3.19, and each recorded day keeps the version it was filed under. NFL: Rule v1.9. NCAAB: Other-Sports Rule v2.1. Pick of the Day: criteria v3.1. The Syndicate: Spec v4.

Sports record reset — 2026-11-18. The Council closed the college football spread book and the player props books, and moved the NBA spread book to a shadow ledger that publishes nothing. College basketball spread plays now file only when the model's projected margin is 6 points or more. All earlier sports results were deleted; the sports record on this page starts with the plays filed on 2026-11-18. The racing record is unchanged.

Series starts. Racing books — the racing pick book, Top Four, the racing Pick of the Day and the Syndicate racing books: November 17, 2026. Sports books — NFL, NCAAB, the sports Pick of the Day cards and the Syndicate sports books: November 18, 2026, the reset date above. Rows filed before a book's start belong to the closed era summarized on the fresh-start page, which stays linked here permanently as part of the record: why this record begins November 17, 2026.

Past performance is not indicative of future results.

Tier 2 — research

Backtests and closed series, kept for reference. These are reconstructed or closed numbers: they are not part of the recorded totals above and never merge into them. Every figure comes from the study's own record file, named in the methods block.

Backtest 1 — the cleaned subset (882 fair races, then 642)Not supported · n = 882

The clean-race study asked whether the speed figure ranks better once racing's noise is removed. The funnel, from the benchmark population of 11,624 races: only fast-dirt and firm-turf races in chart-agreed conditions passed (7,231); races with contender trouble or a did-not-finish in the chart comments fell out (5,143 left); any race with a scratched starter fell out (2,350 left); and the winner's figure had to run within three-quarters of a point of its cell par (882 left). The yield was 882 races, not the round thousand the exercise was named for — no filter was loosened to pad the count.

Arm (882 races)Top-pick win %ROI ($2 win)
Morning line32.99% (291/882)−19.4%
Frozen blend33.45% (295/882)−19.2%
Figure — best last23.81% (210/882)−20.9%
Figure — best median20.38% (171/839)−24.3%

With the noise removed, the figure still does not out-rank the morning line — best-last wins 23.8% against the line's 33.0%. The blend's log-loss edge over the morning line grows slightly on clean ground (0.0391), while its money edge vanishes: both arms lose about 19%. On favorites, the morning line underrates its own prices in fair races — favorites won 33.0% against a 26.8% morning-line-implied rate (+6.2 points) — yet a bet on every favorite still returned −19.4% at final odds, because the closing market captures more than the morning line leaves. The sharpest cells were starter-class favorites (+14.7 points, 72 races) and firm-turf favorites (+13.2 points, 78 races, +15.6% ROI — a hypothesis filed for full-population testing, not a result, at that count).

The amendment (642 races). Races where any starter had no prior speed figure were then removed — 240 races — under a pre-race coverage rule (a starter's figure count is fixed before the race; the rule removes races the pick won and lost alike). On the 642: morning line 33.02% (−17.8%), blend 33.49% (−18.2%), best-last figure 23.99% (−26.1%), best-median figure 19.43% (−35.3%). The blend's log-loss edge widened to 0.0480; the outside-four rate barely moved (18.9% to 18.1%; 167 misses to 116). No conclusion changed.

Backtest 2 — the entire population (11,624 races)Not supported · family closed · n = 11,624

The mission benchmark: pooled walk-forward scoring on 11,624 races and 84,417 starters, every race the frozen corpus holds from March 2023 on. Both faces of the old record lose in every fold, class, surface and field-size band.

ArmLog lossConcordanceTop-pick win %ROI
Morning line (ML-BASE)1.72960.713133.51%−17.80%
Frozen blend1.69890.716533.64%−17.67%

The blend's probabilities are better calibrated than the morning line's (log loss 1.6989 against 1.7296; a uniform guess scores 1.9507) and its top pick wins no more often. Nothing market-free came close: a logistic model scored 1.7693, a PLS model 1.7704 and a gradient-boosted model 1.7791 on the same features and folds — the family is closed. Two from-scratch candidates built for the mission also failed: opponent-adjusted dynamic ratings (log loss 1.9772, top pick 23.07%) and an independent-figure trajectory model (1.9871, 21.97%) — both carry real ranking signal (concordance about 0.60) and both fail on calibration, scoring below the uniform floor. The one anchor that beat both controls was the closing market itself: final odds won 38.7% at −14.3% ROI over 13,459 races — and still lost money; no anchor was frozen or adopted.

Pace, tested three ways on the same population, does not select: re-ranking by pace won 26.23% against the control's 33.51%; pace as a veto replaced picks with horses that won 18.77% against the struck picks' 28.15%; pace-scenario narrowing held the winner 67.10% / 66.91% of the time against the blend top-two control's 67.22%. The speed figure itself is a real measurement with weak precision: start-to-next-start repeatability 0.498, below raw final time's 0.528 on identical pairs; class separation works (winners' medians: stakes 2.00, allowance 1.46, claiming 0.43, maiden special weight 0.14, maiden claiming −0.82); head-to-head margin calls hit 58.2% against the morning line's 65.6%. It stays a blend input and an annotation, never a standalone ranking.

Backtest 3 — sports (four leagues)Not supported · largest population n = 15,337

Holdout populations: NFL 641 games, NBA 1,413, NCAAF 1,827, NCAAB 15,337. Note the scale honestly: the football samples are small — a few hundred games read as a measuring stick, not a verdict — and NCAAB is the only large population of the four.

In the ratings experiment (R1), the de-vigged closing market finished first in every league on holdout log loss: NFL 0.6079, NCAAF 0.5445, NCAAB 0.5388, NBA 0.5745. One candidate earned a track: Glicko-2 in the NFL (0.6317, against the market's 0.6079). Elo and Glicko were rejected in the other three leagues. The lab's own opponent-adjusted differential was the best independent probability source in three of four leagues, and still lost to the market in all four. A performance-inputs model (R2) was not supported on its NFL arm and not run elsewhere; a market blend (R6) failed its gate in every league — the fitted blend weight hit the floor, importing the market rather than beating it.

The Power Ratings v2 verification reads the same way: a real measuring stick with weak precision. Team separation is strong in all four leagues (Spearman against realized ordering: NFL 0.899, NBA 0.910, NCAAF 0.755, NCAAB 0.742), and the ratings agree with published systems where a comparison exists (NCAAF against SP+ at 0.955; NCAAB against Torvik at 0.772). But margin prediction finishes second to the closing spread in every league, and calibration fails its frozen bar in football (error 0.0689 against the market's 0.0365 in the NFL; 0.0388 against 0.0245 in NCAAF; basketball passes). The opponent-adjustment layer improves repeatability in the NFL, NBA and NCAAF and hurts it in NCAAB (0.615 adjusted against 0.656 raw). On clean games — no early- season no-form teams, no tournament back-to-backs, no neutral sites — only NCAAB's model-versus-market gap shrinks (0.0261 to 0.0154). Five candidates and one enabler were filed from this research (noise concentration, the NCAAB adjustment question, football calibration, totals blowout losses, NBA three-point variance, NFL weather data); none is adopted.

Closed series — the §26 consensus watchClosed · n = 559

Under Selection Protocols v3.12–v3.18, a five-factor consensus scored the morning-line top four in every race and its first horse was the booked selection. The machinery was removed in v3.19 — the consensus added coverage, not edge, and compounded the blend's own ranking back onto itself. Its top-four watch closed at 452 of 582 winners in the four (77.7%), ROI -21.4% on the $2 win book. The series stands as its own record and never merges with the Top Four series in Tier 1, which is a different mechanism (the four highest filed blend win probabilities).